Claude Skill MartBrowse skillsQuick linesLearn by videoTerminal guideWhat is a Skill?
 
← Back to list

Agent Harness

Compiles a goal into a verifiable task plan and drives a bounded agent loop with machine-run checks, retry caps, and human escalation.

AutomationAdvanced★ 26,260⑂ 3,699AI score 9/10Last updated: Aug 30, 2026

What it does

  • Compiles a goal into a verifiable plan (plan.json); vague goals are refused (exit 3) with forcing questions instead of guesswork.
  • Selects a per-domain manifest (18 committed domains) describing which skills, tools, and checks actually exist.
  • A loop controller runs execute → verify one task at a time and executes each check itself via subprocess with timeouts and an evidence log — the agent cannot declare its own success.
  • Budgets are terminal: 3 attempts per task, 12 loop iterations by default; exhaustion escalates to a named human instead of faking a pass.
  • All state lives in .agent-harness/state.json, so a fresh session can resume from the plan + state alone.

Who it's for

  • Developers and tech leads who delegate long tasks to agents but want "done" to be provable
  • Anyone chaining multiple skills into a resumable, state-driven automation pipeline
  • Teams that need retry limits, approval gates, and audit trails around agent work

Examples

  1. Push "audit the payments service and design an SLO with an error budget" through the engineering harness until every task verifies
  2. Set up a marketing-domain loop: produce campaign artifacts → run checks → close only when clean
  3. Run loop_controller.py --sample to watch a failed verify consume an attempt and see close refused until evidence-backed success

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/alirezarezvani/claude-skills/HEAD/.gemini/skills/agent-harness/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the .gemini/skills/agent-harness folder from the GitHub repo alirezarezvani/claude-skills into my ~/.claude/skills/agent-harness/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/alirezarezvani/claude-skills.git && mkdir -p ~/.claude/skills/agent-harness && cp -r claude-skills/.gemini/skills/agent-harness/* ~/.claude/skills/agent-harness/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal and clone the repo: git clone https://github.com/alirezarezvani/claude-skills.git
  2. Create the skill folder: mkdir -p ~/.claude/skills/agent-harness
  3. Copy the skill in: cp -r claude-skills/.gemini/skills/agent-harness/* ~/.claude/skills/agent-harness/
  4. Confirm the scripts/, assets/harnesses/, and references/ subfolders came along — the loop won't run without them.
  5. Make sure Python 3 is available: python3 --version
  6. Smoke test: python3 ~/.claude/skills/agent-harness/scripts/loop_controller.py --sample should exit 0.
  7. Restart Claude Code and ask something like "run this goal through the agent harness" to trigger the skill.
View source on GitHub ↗License: MIT